Model selection and sensitivity analysis in the biomechanics of soft tissues: a case study on the human knee meniscus

Elsiddig Elmukashfi, Gregorio Marchiori, Matteo Berni, Giorgio Cassiolas, Nicola Francesco Lopomo, Hussein Rappel, Mark Girolami, Olga BarreraarXiv 2021 · DOI · arXiv

Modeling the human body is a bit like trying to simulate a machine where every part is made of a different, slightly leaky sponge. In the world of biomechanics, the human knee meniscus—a crescent-shaped fibrocartilaginous tissue—is one of the most challenging "sponges" to get right. A recent paper by Elmukashfi et al. provides a comprehensive roadmap for navigating this complexity, comparing various mathematical models and advocating for a shift toward probabilistic, Bayesian approaches.

TL;DR

This paper evaluates how different material models (linear vs. nonlinear, isotropic vs. anisotropic, and time-dependent) impact the simulated mechanical behavior of the human knee meniscus. The authors demonstrate that accounting for anisotropy and viscoelasticity is crucial for accuracy and propose Bayesian inference as the gold standard for handling the high variability and scarcity of biological data.

What Problem Does This Paper Address?

The meniscus acts as the knee's shock absorber, load distributor, and stabilizer. It is a complex, "biphasic" material consisting of a solid collagen matrix and a fluid phase (mostly water). Because its internal structure is heterogeneous and its properties vary depending on the direction of force (anisotropy), modeling it accurately is notoriously difficult.

The core challenge is the paucity and variability of data. Experimental tissue testing is non-standardized, and results vary wildly between individuals. Traditional "deterministic" models—which provide a single best-guess value for material parameters—fail to account for this uncertainty. This paper seeks to determine which modeling "ingredients" are essential for a realistic simulation and how to better handle the "messiness" of biological data.

Methods

The researchers combined a rigorous systematic literature review with advanced computational modeling:

  1. Systematic Review: They analyzed 24 studies on human meniscus poro-mechanics to synthesize the current state of experimental data.
  2. Finite Element (FE) Modeling: They developed a patient-specific 3D model of a knee joint (including bones, cartilage, and menisci) using MRI data.
  3. Model Comparison: They tested a hierarchy of constitutive equations:
  4. Linear Elastic: The simplest, assuming a constant stiffness.
  5. Hyperelastic (Isotropic): Nonlinear models like Neo-Hookean and Ogden that account for large deformations.
  6. Anisotropic (HGO): The Holzapfel-Gasser-Ogden model, which accounts for the specific orientation of collagen fibers.
  7. Viscoelastic: Models that account for time-dependent behavior (how the tissue reacts to the speed of loading).
  8. Bayesian Framework: The authors presented a theoretical outlook on using Bayesian inference to treat material parameters as probability distributions rather than fixed numbers.

Datasets

  • Experimental Stress-Strain Data: Primarily sourced from Proctor et al. (1989), featuring uniaxial tensile tests on bovine meniscus specimens.
  • Literature Synthesis: A meta-analysis of 24 studies provided a range of values for the equilibrium modulus ($E_{eq}$), permeability ($k$), and fiber modulus ($E_f$).
  • Anatomical Data: A subject-specific knee geometry segmented from MRI sagittal proton density sequences.

Key Insights and Ideas

  • Linearity is a Trap: Linear elastic models are common because they are computationally "cheap," but they are far too stiff at low strains. They overestimated contact pressure by 5.3 MPa compared to nonlinear Ogden models.
  • Anisotropy is the "Softener": When the model accounts for the fact that collagen fibers are aligned circumferentially (anisotropy), the overall response of the tissue becomes "softer" and more realistic. This is because the model can finally account for the "crosstalk" between radial and circumferential deformations.
  • The Speed of Life Matters: Soft tissues are viscoelastic. During dynamic activities like power walking, the "time-dependent" response makes the tissue appear much stiffer due to viscous dissipation. The difference in contact pressure between time-independent and time-dependent models was a staggering 20 MPa.
  • From Points to Probabilities: The authors argue that because biological data is so scarce, we should stop asking "What is the value of Young’s Modulus?" and start asking "What is the probability distribution of this modulus?" Bayesian inference allows researchers to incorporate "prior" knowledge and produce "credible intervals" for their predictions.

Key Results

  • Best Fit: The Ogden model (with two parameters) provided the best fit for isotropic experimental data, with a root-mean-square error of only ~0.93 MPa.
  • Pressure Spikes: In a simulated power-walking scenario, the peak contact pressure on the tibial cartilage reached 51.23 MPa when using a viscoelastic anisotropic model, significantly higher than the ~31 MPa predicted by time-independent models.
  • Deformation Patterns: The lateral meniscus showed greater and more irregular movement than the medial meniscus, which is more constrained by its attachments to the tibial plateau.

Limitations and Future Work

The authors acknowledge that while they used a patient-specific geometry, the material parameters were largely derived from bovine data due to the lack of comprehensive human datasets. Furthermore, the model simplified the "meniscal horns" (the attachment points) as being fixed to the bone, whereas in reality, these attachments have their own complex compliance.

Future work points toward "on the fly" data integration. Imagine a surgeon using interventional medical imaging to acquire a patient's specific tissue properties during a procedure, feeding that data into a Bayesian model to predict the optimal surgical intervention in real-time.

Reference: Elmukashfi, E., et al. (2021). Model selection and sensitivity analysis in the biomechanics of soft tissues: a case study on the human knee meniscus. arXiv:2112.13434v1.

Summary updated

Abstract

Soft tissues - such as ligaments and tendons - primarily consist of solid (collagen, predominantly) and liquid phases. Understanding the interaction between such components and how they change under physiological loading sets the basis for elucidating the essential link between their internal structure and mechanical behaviour. In fact, the internal heterogeneous structure of this kind of tissues leads to a wide range of mechanical behaviours, which then determine their own function(s). Characterising these behaviours implies an important experimental effort in terms of tissue harvesting, samples preparation and implementation of testing protocols - which, often, are not standardised. These issues lead to several difficulties in both collecting and providing comparable and reliable information. In order to model the behaviours of heterogeneous tissues and identify material parameters, a large volume of reproducible experimental data is required; unfortunately, such an amount of information is often not available. In reality, most of the studies that are focused on the identification of material parameters, are largely based on small sets of experimental data, which present a large variability. Such a large variability opens on to uncertainties in the estimation of material parameters, as reported in the literature. Hence, the use of a rigorous probabilistic framework, that is able to address uncertainties due to paucity of data, is of paramount importance in the field of biomechanics; in this perspective, Bayesian inference represents a promising approach. This study was focused on the analysis of the knee meniscus as a paradigmatic example of human soft tissue.
BibTeX
@misc{elmukashfi2021model,
  author = {Elsiddig Elmukashfi and Gregorio Marchiori and Matteo Berni and Giorgio Cassiolas and Nicola Francesco Lopomo and Hussein Rappel and Mark Girolami and Olga Barrera},
  title = {{Model selection and sensitivity analysis in the biomechanics of soft tissues: a case study on the human knee meniscus}},
  year = {2021},
  eprint = {2112.13434},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2112.13434},
}
Figure